#16298: feat(xai): switch grok-4-1-fast variants by thinking level

## Summary - Add thinking-aware model-family resolution for xAI `grok-4-1-fast` so runtime routing selects `grok-4-1-fast-reasoning` for thinking-on levels and `grok-4-1-fast-non-reasoning` for `/think off`. - Integrate the resolver into reply/runtime session model resolution and status output so effective model selection is consistent across execution paths. - Add targeted coverage for family detection, runtime routing behavior, and session model resolution fallbacks. ## Verification - `pnpm test -- src/agents/model-selection-thinking.test.ts src/auto-reply/reply/model-selection.override-respected.test.ts src/gateway/session-utils.model-ref.test.ts` - `pnpm tsgo` - `pnpm build` <!-- greptile_comment --> <h3>Greptile Summary</h3> Added thinking-aware model resolution for xAI `grok-4-1-fast` that automatically switches between `-reasoning` and `-non-reasoning` variants based on thinking level (`/think off` vs thinking-on). - Introduced `model-families.ts` to define reasoning model families and `isReasoningFamilyAllowed` to check allowlist compatibility - `resolveThinkingAwareModelRef` in `model-selection-thinking.ts` handles runtime routing - Integrated into reply execution (`get-reply-run.ts`), model selection state (`model-selection.ts`), session resolution (`session-utils.ts`), and status output (`commands-status.ts`) - Added comprehensive test coverage across family detection, runtime routing, session model resolution, and allowlist behavior The implementation follows existing patterns and provides good test coverage. One minor issue: the status command doesn't pass `allowedModelKeys` to `resolveThinkingAwareModelRef`, which could cause it to display a model variant that would be blocked by allowlist restrictions (though this is an edge case). <h3>Confidence Score: 4/5</h3> - This PR is safe to merge with one minor edge case to address - The implementation is well-structured with comprehensive test coverage, follows existing codebase patterns, and correctly handles the core functionality. Score is 4 (not 5) due to the missing `allowedModelKeys` parameter in the status command's thinking-aware resolution, which could cause status output to show an incorrect model variant in edge cases involving allowlists. This is a display-only issue that wouldn't affect actual execution. - src/auto-reply/reply/commands-status.ts requires attention for allowlist handling _{Last reviewed commit: 07f51c4} <!-- greptile_other_comments_section --> _{(4/5) You can add custom instructions or style guidelines for the agent [here](https://app.greptile.com/review/github)!} <!-- /greptile_comment -->